That’s why I searched another fit. This fit both produces smaller residuals and  more normal residual distribution. However, the downside is it is more complex in parameter-wise comparing to original method. Although the R^2 prediction is very good the more terms brings more risk of misfit in the extrapolations.